PEP: 101
Title: Doing Python Releases 101
Version: $Revision$
Last-Modified: $Date$
Author: barry@zope.com (Barry A. Warsaw), guido@python.org (Guido van Rossum)
Status: Active
Type: Informational
Created: 22-Aug-2001
Post-History:


Abstract

    Making a Python release is an arduous processes that takes a
    minimum of half a day's work even for an experienced releaser.
    Until recently, most -- if not all -- of that burden was borne by
    Guido himself.  But several recent releases have been performed by
    other folks, so this PEP attempts to collect, in one place, all
    the steps needed to make a Python release.

    (Note: These steps were recorded during the Python 2.2a2 release
    done by Barry and Guido.  It will be checked for accuracy with the
    subsequent release, to be done by Barry.)


How to Make A Release

    Here are the steps taken to make a Python release.  Some steps are
    more fuzzy than others because there's little that can be
    automated (e.g. writing the NEWS entries).  Where a step is
    usually performed by An Expert, the name of that expert is given.
    Otherwise, assume the step is done by the Release Manager (RM),
    the designated person performing the release.  Almost every place
    the RM is mentioned below, this step can also be done by Guido,
    naturally!

    XXX: We should include a dependency graph to illustrate the steps
    that can be taken in parallel, or those that depend on other
    steps.

    We use the following conventions in the examples below.  Where a
    release number is given, it is of the form X.YaZ, e.g. 2.1a3 for
    Python 2.1 alpha 3, where "a" == alpha, "b" == beta, "rc" ==
    release candidate, and "f" == final.  If a micro release number is
    used, then we'll say X.Y.MaZ.

    - A day or two before the release, create a branch for X.YaZ.

      All Python development happens on the trunk.  A few days before
      the release, the RM will create a branch for this release.  Only
      the RM is authorized to make commits to this branch, but
      check ins can proceed as normal on the trunk.  It is the
      responsibility of the RM to decide on a case-by-case basis which
      trunk modifications should be merged into the branch.

      To create a branch the following steps are taken:

      * Do a CVS update with the -A flag, e.g.
        % cvs update -A

      * CVS tag the trunk with the symbolic name "rXYaZ-fork", e.g.
        % cvs tag r22a3-fork

      * Make the branch with the symbolic name "rXYaZ-branch", e.g.
        % cvs tag -b r22a3-branch

      * Check out a clean version of the branch into a new directory.
        You'll be doing a lot of work in this directory and you want
        to keep it straight from your trunk working directory.  E.g.
        % cvs -d <cvsroot> -q co -d python-22a3 -r 22a3-branch python/dist/src

    - Send an email to python-dev@python.org indicating the fork and
      branch tags you've just created.

    - Put a freeze on check ins into the branch.  At this point,
      nobody except the RM should make any commits to the branch (or
      his duly assigned agents, i.e. Guido the BDFL, Fred Drake for
      documentation, or Tim Peters for Windows).  If the RM screwed up
      and some desperate last minute change to the branch is
      necessary, it can mean extra work for Fred and Tim.  So try to
      avoid this!

    - In the branch, change Include/patchlevel.h in two places, to
      reflect the new version number you've just created.  You'll want
      to change the PY_VERSION macro, and one or several of the
      version subpart macros just above PY_VERSION, as appropriate.

    - For the next few days, selectively merge stuff from trunk into
      branch.  For each change you see on the trunk (i.e. via the
      python-checkins mailing list), you need to decide whether the
      change should also be applied to the branch.

      If so, it's fairly easy to apply the change by diff'ing the file
      and patching it manually.  You can also sometimes get away with
      just copying the file from the trunk directory to the branch
      directory, but be careful so you don't lose changes that only
      exist in the branch!

    - The night before the release, the RM updates the Misc/NEWS file,
      adding high level items new to this release.  E.g. if we're
      releasing 2.2a3, there must be a section at the top of the file
      explaining "What's new in Python 2.2a3".  It will be followed by
      a section entitled "What's new in Python 2.2a2".

      Note that you /hope/ that as developers add new features to the
      trunk, they've updated the NEWS file accordingly.  You can't be
      positive, so double check.  If you're a Unix weenie, it helps to
      verify with Tim Peters about changes on Windows, and Jack Jansen
      about changes on the Mac.

    - Once the branch is frozen, Fred Drake needs to create the HTML
      from the documentation.  He does this and uploads the file to
      www.python.org.  Then he tells Tim Peters where this file is.
      This may generate some last minute changes on the branch.

    - Tim Peters grabs the HTML and uses this to build the Windows
      installer.  Tim then creates a new "release" named X.YaZ on the
      SourceForge file release manager.

      (Diversion: SF's file manager has "packages" and "releases".  We
      use packages to name major upcoming releases, e.g. python-2.2 or
      python-2.1.1.  Inside each package are a number of "releases"
      for each new actual release -- i.e. the thing you're building.
      An example of a release name is 2.2a3.  Once created, packages
      and releases are never deleted, but old ones are hidden to
      reduce confusion.  More on this below.)

      If this is the first release for this major Python version, Tim
      will create a new package containing the major Python version
      number.

    - Tim performs his Windows magic, generating an installer
      executable.  He uploads this file to SourceForge under the
      release he just created.  He then sends the RM a notice which
      includes the MD5 checksum of the Windows executable.

      Note that Tim's creation of the Windows executable may generate
      a few more commits on the branch.

    - Now, you're ready to build the source tarball.  First cd to your
      working directory for the branch.  E.g.
      % cd .../python-22a3

    - Do a "cvs update" in this directory.  You should not see any "M"
      files, but you may see several "P" files.  I.e. you better not
      have any uncommitted changes in your working directory, but you
      may pick up some of Fred's or Tim's last minute changes.

    - Now tag the branch using a symbolic name like "rXYaZ",
      e.g. r22a3
      % cvs tag r22a3

    - Change to a neutral directory, i.e. one in which you can do a
      fresh, virgin, export of the branch.  You will be creating a new
      directory at this location, to be named "Python-X.YaZ".  Do a
      CVS export of the tagged branch.
      % cd ~
      % cvs -d <cvsroot> export -rr22a3 -d Python-2.2a3 python/dist/src

    - Generate the tarball.  Note that we're not using the `z' option
      on the tar command because 1) that's only supported by GNU tar
      as far as we know; 2) we're going to max out the compression
      level, which isn't a supported option.
      % tar cf - Python-2.2a2 | gzip -9 > Python-2.2a2.tgz

    - Calculate the MD5 checksum of the tgz file you just created
      % md5sum Python-2.2a2.tgz

      Note that if you don't have the md5sum program, there is a
      Python replacement in the Tools/scripts/md5sum.py file.

    - Now you want to perform the very important step of checking the
      tarball you just created, to make sure a completely clean,
      virgin build passes the regression test.  Here are the best
      steps to take:

      % cd /tmp
      % tar zxvf ~/Python-2.2a3.tgz
      % cd Python-2.2a3
      % ls
      (Do things look reasonable?)
      % ./configure
      (Loads of configure output)
      % make test
      (Do all the expected tests pass?)

      If the tests pass, then you can feel good that the tarball is
      fine.  If some of the tests pass, or anything else about the
      freshly unpacked directory looks weird, you better stop now and
      figure out what the problem is.

    - Start your upload to SF.  You need to get Python-2.2a3.tgz into
      SourceForge.  This can take a while both because of the time it
      takes to upload such a huge file, /and/ because SF has a 30
      minute delay built into the file release process.  The next few
      steps can be taken in parallel, so it's best to start the upload
      now and keep an eye on its progress.

      I've found that the `ncftpput' program is a great tool to use if
      you have it available.  You can execute the following command to
      do the upload:
      % ncftpput -F upload.sf.net incoming Python-2.2a3.tgz

      If you don't have ncftpput around, you can use whatever ftp
      client you're comfortable with.  Just be sure that you're
      uploading this to the "incoming" directory on upload.sf.net.

    - You also need to upload the tgz file to creosote.python.org.
      Usually Tim will have already uploaded the exe file to creosote,
      but if not, you'll need to do that too.  These steps can take a
      long time depending on your network bandwidth.  You have two
      choices:

      1) Upload them to SF first, then wget them from creosote.  Pros:
         easy to do; much friendlier to your own personal bandwidth.
         Cons: can take even longer because you're subject to the 30
         minute SF file upload delay, and the upload rate from
         SF->creosote never seems to get above 20 KB/sec.

      2) scp both files from your own machine to creosote.  Pros: you
         avoid the 30 minute SF delay.  Cons: you don't get much else
         done if you're on a small pipe.

    - While you're waiting, you can start twiddling the web pages to
      include the announcement.

      * In the python.org web site CVS tree, cd to the X.Y
        subdirectory, and copy index.ht to new-index.ht

        % cd .../pydotorg/2.2
        % cp index.ht new-index.ht

      * Edit the file for content: usually you can globally replace
        X.Ya(Z-1) with X.YaZ.  However, you'll need to think about the
        "What's New?" section.  You also need to watch out about two
        specific links in the file, one which references "the full
        scoop" on changes -- a shownotes link, and one which
        references the downloads on SF -- a showfiles link.  You won't
        be able to update either link until you've completed the steps
        below.  For now just note what needs to change.

        Also, update the MD5 checksums.

      * Preview the web page by doing a "make" -- NOT a "make install".
        View the page via a file: url.

      * Similarly, edit the ../index.ht file, i.e. the python.org home
        page.  In the Big Blue Announcement Block, move the paragraph
        for the new version up to the top and boldify the phrase
        "Python X.YaZ is out".  Edit for content, and preview as
        above.  Do NOT do a "make install" yet!

    - Now we're waiting for the ncftpput command, and the scp to
      creosote to finish.  Da de da, da de dum, hmm, hmm, dum de dum.

    - Do the SourceForge file release dance.

      * Go to the Python project and click on "Admin"
      * Click on "Edit/Release Files"
      * Since Tim has usually by now created the package and release
        we're going to use, scroll down and click on "Edit Releases"
        for the package we're releasing into.
      * Find the release named X.YaZ and click on "Edit This Release"

      You should now perform Step 1 of the file release dance...

      * The "Status" field should be "Active" not "Hidden"
      * In the text box that says "Paste The Notes In", paste in all
        the "What's New" entries from the Misc/NEWS file that describe
        this major version of Python, /not/ just the ones for this
        particular release.  E.g. If we're releasing Python 2.2a3,
        we'd include the "What's New" sections for Python 2.2a3,
        2.2a2, and 2.2a1.
      * Leave the "Paste The Change Log In" section blank, but click
        on "Preserve my pre-formatted text".
      * Hit the Submit/Refresh button for Step 1.

      This will bring you back to the file release page.  DO NOT do
      the following step until your ftp upload is complete!  Once it
      is, you can perform Step 2 of the file release dance...

      * Click on the checkbox next to the file Python-X.YaZ.tgz.  Be
        sure no other box is checked!  Then click on the "Add Files
        and/or Refresh View" button for Step 2.

      And now, Step 3...

      * There should be exactly two files listed here, one is the tgz
        file you just added, and the other is the exe file that Tim
        added earlier.
      * For the tgz file, be sure that the "Processor" field says
        "Any" and the "File Type" field says "Source .gz".
      * Click on "Update/Refresh" for the .tgz file.
      * For the exe file, make sure that the "Processor" field says
        "i386" and the "File Type" field says "Other".  Tim usually
        gets this right <wink>, but if not, make any necessary changes
        and click on "Update/Refresh" for the exe file.

     Step 4...

     DO NOT DO STEP 4 NOW.  Wait until after you send out the email
     announcement to send the SF email notice.

    - Still on SF, click on the "Files" button at the top of the
      page.  Find the release you've just made and click on it -- not
      on the tgz or exe file, but on the release link under the
      package name.  E.g. package named python-2.2, click on the
      "2.2a3" link.

      This should be a page that says "Release Name: X.YaZ" and it
      should contain the "What's New" sections you pasted in earlier.
      Note the url of this page.  Copy and paste it into the
      pydotorg/X.Y/new-index.ht file you created above.  This is the
      "shownotes" link mentioned earlier.

    - Now click on the "Summary" link at the top of the page and
      scroll down to the "Latest File Releases" section.  Find the
      package you just made a release for (the Version should be
      X.YaZ, and the Date should be today's date).  Click on the
      "Download" link.

      Your new release should be highlighted in pink.  Note the url
      for this page.  Copy and paste it into the
      pydotorg/X.Y/new-index.ht file from above.  This is the
      "showfiles" link mentioned earlier.

    - Now you need to go to creosote.python.org and move all the files
      in place over there.  Our policy is that every Python version
      gets its own directory, but each directory may contain several
      releases.  We keep all old releases, moving them into a "prev"
      subdirectory when we have a new release.

      So, there's a directory called "2.2" which contains
      Python-2.2a2.exe and Python-2.2a2.tgz, along with a "prev"
      subdirectory containing Python-2.2a1.exe and Python-2.2a1.tgz.

      So...

      * On creosote, cd to ~ftp/pub/python/X.Y creating it if
        necessary.

      * Move the previous release files to a directory called "prev"
        creating the directory if necessary (make sure the directory
        has g+ws bits on).  If this is the first alpha release of a
        new Python version, skip this step.

      * Move the .tgz file and the .exe file to this directory.  Make
        sure they are world readable.  They should also be group
        writable, and group-owned by webmaster.

    - Update the X.Y/bugs.ht file if necessary.  You may need to get
      BDFL input for this step.

    - Now preview the new-index.ht file once more.  IMPORTANT: follow
      every link on the page to make sure it goes where you expect it
      to go, and that what you expect to be there is there.

    - If everything looks good, move new-index.ht to index.ht and do a
      "make install" in this directory.  Go up to the parent directory
      (i.e. the root of the web page hierarchy) and do a "make
      install" there too.  You're release is now live!

    - Now it's time to write the announcement for the mailing lists.
      This is the fuzzy bit because not much can be automated.  You
      can use one of Guido's earlier announcements as a template, but
      please edit it for content!

      Once the announcement is ready, send it to the following
      addresses:

      python-list@python.org
      python-announce@python.org
      python-dev@python.org

    - Go back to the file releases page on SF and complete Step 4,
      sending out the email notification.

    Now it's time to do some cleanup.  These steps are very important!

    - Go back to SF, Admin->Edit/Release Files.  Click on "Edit
      Releases" for the package you just added to.  For each old
      release, click on "Edit This Release" and under Step 1, change
      the "Status" to "Hidden".  Click on the Step 1 Submit/Refresh
      button.

    - Merge the branch back into the trunk!  Now that we've released
      this branch, we don't need it any more.  We've already tagged it
      so we can always reproduce it.  Note that merging branches is a
      bit of a black art, but here's what's worked for us.

      * Check out a completely clean, virgin working directory of the
        trunk, by doing this in the directory that is the parent of
        your branch working directory python-XYaZ:
        % cvs -d <cvsroot> co -d python-clean python/dist/src

      * Run a diff against your branch by doing this in the common
        parent directory containing both python-clean and python-XYaZ:
        % diff -r python-clean python-22a2 | grep ^diff | grep -v CVS

      * Take the output of this and stick it in a file,
        e.g. /tmp/diffcmd.sh

      * Edit diffcmd.sh to get rid of files that you know don't have
        important changes.  You're looking for files that have updates
        in the branch that haven't made it to the trunk.  If you've
        been diligent about merging changes from the trunk into the
        branch, there shouldn't be many of these files.

      * Edit /tmp/diffcmd.sh, changing all the -r's into -u's.  Run
        the /tmp/diffcmd.sh command like so:
        % sh /tmp/diffcmd.sh > /tmp/pydiff.txt

      * Attempt to patch your python-clean working directory.  Do this
        first, noting that --dry-run does not actually apply any
        patches, it just makes sure that the patch command runs
        successfully to completion:
        % patch -p1 --dry-run < /tmp/pydiff.txt

      * If this goes well, run it again, taking out the --dry-run
        option.  If this fails, or if it prompts you for a file to
        patch, try using -p0 instead of -p1.  Otherwise, your diff
        command was messed up, so try again.

      * cd to python-clean and do a "cvs commit".  Use as your log
        message something like "Merging the rXYaZ-branch tag back into
        the trunk".

      * Edit the file Include/patchlevel.h so that the PY_VERSION
        string says something like "X.YaZ+".  Note the trailing `+'
        indicating that the trunk is going to be moving forward with
        development.  E.g. the line should look like:

        #define PY_VERSION              "2.2a2+"

        Commit this change.

      * Now test your clean, merged trunk by doing
        % make distclean
        % ./configure
        % make test

        To ensure that the regression test suite passes.  If not, you
        screwed up somewhere!

    Step 5 ...

    Verify!  This can be interleaved with Step 4.  Pretend you're a
    user:  download the files from python.org *and* SourceForge, and make
    Pythons from them.  This step is too easy to overlook, and on
    several occasions we've had useless release files.  Once a general
    server problem caused mysterious corruption of all files; once the
    source tarball got built incorrectly; more than once the file upload
    process on SF truncated files; and so on.


What Next?

    Rejoice.  Drink.  Be Merry.  Write a PEP like this one.  Or be
    like unto Guido and take A Vacation.

    You've just made a Python release!


Windows Notes

    Windows has a GUI installer, various flavors of Windows have "special
    limitations", and the Windows installer also packs precompiled
    "foreign" binaries (Tcl/Tk, expat, etc).  So Windows testing is
    tiresome but very necessary.

    Concurrent with uploading the installer, I install Python from it
    twice:  once into the default directory suggested by the installer,
    and later into a directory with embedded spaces in its name.  For
    each installation, run the full regression suite from a DOS box, and
    both with and without -0.  Also try *every* shortcut created under
    Start -> Menu -> the Python group.  When trying IDLE this way, verify
    that Help -> Python Documentation works.  When trying pydoc this way
    (the "Module Docs" Start menu entry), make sure the "Start Browser"
    button works, and make sure you can search for a random module (I
    use "random" <wink>) and then that the "go to selected" button works.
    It's amazing how much can go wrong here -- and even more amazing how
    often last-second checkins break one of these things.  If you're "the
    Windows geek", keep in mind that you're likely the only person
    routinely testing on Windows, and that Windows is simply a mess.

    Repeat all of the above on at least one flavor of Win9x, and one of
    NT/2000.  On NT/2000, try both an Admin and a plain User (not Power
    User) account.

    WRT Step 5 above (verify the release media), since by the time release
    files are ready to download I've generally run many Windows tests on
    the installer I uploaded, I usually don't do anything for Step 5
    except a full byte-comparison ("fc /b" if using a Windows shell) of
    the downloaded file against the file I uploaded.


Copyright

    This document has been placed in the public domain.



Local Variables:
mode: indented-text
indent-tabs-mode: nil
End: